Indian Journal of Medical Microbiology Home 

REVIEW ARTICLE
[Download PDF]
Year : 2014  |  Volume : 32  |  Issue : 3  |  Page : 229--235

Meta-analysis in microbiology

N Pabalan1, H Jarjanazi2, TS Steiner3,  
1 School of Natural Sciences and Nursing, Toronto, Ontario, Canada
2 Saint Louis University, Baguio, Philippines, Environmental Monitoring and Reporting Branch,Ontario Ministry of the Environment , Toronto, Ontario, Canada
3 Department of Medicine , Division of Infectious Diseases, University of British Columbia, Vancouver, British Columbia, Canada

Correspondence Address:
H Jarjanazi
Saint Louis University, Baguio, Philippines, Environmental Monitoring and Reporting Branch,Ontario Ministry of the Environment , Toronto, Ontario
Canada

Abstract

The use of meta-analysis in microbiology may facilitate decision-making that impacts public health policy. Directed at clinicians and researchers in microbiology, this review outlines the steps in performing this statistical technique, addresses its biases and describes its value in this discipline. The survey to estimate extent of the use of meta-analyses in microbiology shows the remarkable growth in the use of this research methodology, from a minimal Asian output to a level comparable with those of Europe and North America in the last 7 years.

How to cite this article:
Pabalan N, Jarjanazi H, Steiner T S. Meta-analysis in microbiology.Indian J Med Microbiol 2014;32:229-235

How to cite this URL:
Pabalan N, Jarjanazi H, Steiner T S. Meta-analysis in microbiology. Indian J Med Microbiol [serial online] 2014 [cited 2020 Nov 24 ];32:229-235
Available from: https://www.ijmm.org/text.asp?2014/32/3/229/136547

Full Text

 Introduction



Exponential growth of research outputs makes it increasingly difficult to discern usable knowledge in a flood of information. It is not surprising that primary studies addressing the same issue sometimes produce contradictory results. Conflicting findings are not easy to reconcile and in some cases, an abundance of primary studies often hinders meaningful integration of results using traditional narrative methods. Clearly, it is essential to have some means of systematically comparing and contrasting the results among relevant studies.

Used extensively to compile results of individual studies assessing risk factors in epidemiological studies and effectiveness of healthcare interventions, meta-analysis is useful in decision-making that directs public health policy. Historical application of meta-analysis in public health and microbiology may not be as extensive as in evidence-based medicine. Yet, meta-analysis has addressed key research questions in microbiology such as prevalence of pathogens [1],[2] and prevalence of infectious disease. [3]

This review is directed at laboratory investigators and clinicians for a better understanding of meta-analysis research in microbiology. For purposes of clarifying our points in explaining this research methodology, we use an example from a recent paper that addresses the relationship between Enteroaggregative Escherichia coli (EAEC) colonisation and acute diarrhoea in South-Asian children. [4]

 What is Meta-Analysis?



Meta-analysis is a logically formal and objective technique as well as quantitative mode of summarising research findings, helping to identify genuine associations. Considered at the top of the hierarchy of evidence, [5] this statistical methodology integrates results of independent but related studies to synthesise summaries.

 Why do Meta-Analysis?



The reasons for performing a meta-analysis have to do with sample sizes of the studies, when they are large but results conflict, or when they are small, but their positive findings are not consistent. [6] Primary studies often do not have enough statistical power to assess relationships between risks (interventions) and outcomes. Being most useful when individual studies are too small to yield valid conclusions, meta-analysis increases power, reduces risk of error and facilitates exploratory analysis to generate hypotheses for future research. [7]

 Performing Meta-Analysis



Literature search and data abstraction

A publishable meta-analysis starts with a well-formulated and answerable question considering the time, cost and available resources. In microbiology studies, this includes availability of a good number of primary studies that address associations of key risk factors (e.g. pathogens, genetic susceptibility) with infectious disease. The published Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement recommends that a full electronic search strategy for at least one major database be presented, [8] although such an approach had been deemed insufficient by some. [9] Still a typical search strategy should involve electronic retrieval of all available literature, which includes digital sources such as PubMed (http://www.ncbi.nlm.nih.gov/pubmed/), ScienceDirect (www.sciencedirect.com), Institute of Scientific Information (ISI) Web of Knowledge (http://www.isiwebofknowledge.com) and Google Scholar (http://scholar.google.com). For greater precision in this step, additional measures to exhaustively identify eligible studies include manual searching of relevant journals, references lists and personal contact with researchers. [Figure 1] is a representative flow-diagram depicting the steps taken in searching for relevant literature using key words such as "diarrhoea children Asia" and "Escherichia coli0". It maps out the number of records identified, included and excluded, and the reasons for exclusions. [Table 1] shows the next step which is abstraction of both qualitative (e.g. last name of first author) and quantitative (e.g. sample sizes and statistical power) data from the collection of eligible studies. [4]{Figure 1} {Table 1}

Summary effects calculations

Meta-analysis reports findings in terms of effect sizes, which provides information about how much change is evident across all studies and for subsets of studies. In the meta-analyses example, results of each study are graphically presented in a forest plot [Figure 2] which uses the odds ratio (OR) and 95% confidence interval (CI) metric. The OR has convenient mathematical properties, which allow for ease in combining data and testing the overall effect for significance. [10] The forest plot [Figure 2] in this example, generated from the meta-analysis software, Review Manager (RevMan) is composed of five columns, the leftmost with qualitative data (study, named by last name of first author) and the remaining four with quantitative data. Between columns three and four is the forest plot with a solid vertical line (labelled 1 on the x-axis) which corresponds to the null effect. The area to the left of the vertical line indicates decreased risk and to its right, increased risk. The two leftmost columns show the raw data (cases/controls) from which the OR and 95% CIs (rightmost column) are calculated. In this figure, ORs of the 18 studies are each represented by a blue square and a black horizontal line, representing the point estimate and 95% CIs, respectively. Of the 18 studies, six do not cross the vertical line, two [11],[12] of which are in the decreased risk area and the remaining four [13],[14],[15],[16] in the increased risk area. The 95% CIs of all these studies do not cross 1.0 (null effect) indicating significant effects (P ≤ 0.05) and those of the remaining 12 studies cross the vertical line indicating that the effect estimates were non-significant (P > 0.05). Weight of the study in the meta-analysis is indicated in column four of [Figure 2] presented as proportion of the study that contributes to the pooled OR. The diamond (♦) at the bottom of the forest plot represents the pooled summary effect (OR 1.51, 95% CI 1.12-2.04) which shows that EAEC is significantly associated with acute diarrhoea in Asian children (P = 0.008).{Figure 2}

Modifier analyses

A summary or pooled effect needs to be tested further to ensure rigor of this methodology. In the meta-analysis example, three tests were undertaken; first, subgroup analysis used categories such as geography, span as well as power of the studies and the assays used for case definition [Table 1]. Findings showed that all subgroups had more imprecise ORs compared with the overall pooled OR and that only the statistically powerful studies (>80%) and those that used the "gold standard" HEp2 assay reflected the significant overall effect [Figure 3]a]. Second, using the Galbraith plot method, [17] five outliers were detected, showing three studies above [13],[16],[18] and two below [11],[12] the confidence limits [Figure 4]. The outcome of outlier analysis increased the number of significant ORs, narrowed all CIs, in the overall and subgroups conferring greater precision on the pooled ORs [Figure 3]b].{Figure 3}{Figure 4}

SigmaPlot compared with RevMan

[Figure 3] [Figure 4], [Figure 5] [Figure 6] in this review were generated from SigmaPlot 11.0 which displays the versatility of this software in graphically expressing meta-analysis results. SigmaPlot has a number of advantages over RevMan. (i) In the forest plots, [Figure 3]a and b] SigmaPlot allows control of graphically weighting the point estimates according to sample size (larger squares indic ate larger sample size). (ii) Furthermore, the squares can be filled or not. Filled black squares (■) indicates significance whose CIs do not cross the null effects line, while CIs that do are indicated by unfilled white squares (□) that represent non-significance. (iii) In a simple interface with Microsoft Excel, SigmaPlot provides a graphical summary of heterogeneity [Figure 3].{Figure 5}{Figure 6}

 Biases in Meta-Analysis



Heterogeneity

Heterogeneity is the methodological, epidemiological and clinical dissimilarity across various studies, and meta-analysts spend considerable effort in addressing this issue. When component studies in a meta-analysis are similar to each other, the fixed-effects method of analysis [19] is applied based on the assumption that associations are the same across studies and recognising that the collection of eligible literature is not heterogeneous. When they are not, the random effects analysis model is used, [20] which assumes variability across populations usually resulting in a wider CI. [10]

Statistically, heterogeneity is estimated using a Chi-square-based Q test [21] and quantified with the I 2 metric which shows what proportion of the total variation across studies is beyond chance. [22] Values of I2 lie between 0 and 100% where a value >75% may be considered substantial heterogeneity. [22] In [Figure 2], the Chi-square value (92.1, P < 0.0001) and I 2 value of 82% indicate high heterogeneity.

Graphically, the Galbraith plot is used to detect which component studies (outliers) contribute to heterogeneity. [17] Exclusion of these outliers either reduced or removed heterogeneity of the original findings [Figure 3]a] and is summarised in [Figure 3]b. In [Figure 3]A and 3B, red boxes indicate presence of heterogeneity and green, its absence and show either loss or reduction of heterogeneity with application of outlier analysis.

Publication bias

Publication bias [23] is an issue where significant findings receive priority in published literature over those whose results are non-significant. This bias is evaluated graphically with the funnel plot [Figure 5]. Here, the effect estimate from each study in the meta-analysis is scattered against a measure of its precision, usually 1/SE (standard error). This figure shows a symmetrical distribution of the points with small studies scattered along the length of the x-axis but still centred on the OR estimates from large, more precise studies. This is consistent with absence of bias. To confirm such absence, Egger's regression asymmetry [10] and Begg's and Mazumdar's rank correlation [24] tests were applied. In the meta-analysis example, these tests generated P values of 0.20 and 0.22, respectively, [4] indicating that conclusions were not altered because of this issue. For comparison, [Figure 6] shows a simulated funnel plot indicating presence of publication bias which shows an asymmetrical distribution of the points.

 Value of Meta-Analysis in Microbiology



Interest in the role of pathogens in infectious disease has grown over the past decade. Rapid advancements in identifying key pathogens using molecular techniques have resulted in large amounts of published epidemiological evidence on infection-disease associations. In the meta-analysis example, one aspect of subgroup treatment included the evolving use of assays in identifying EAEC over a period of 22 years (1989-2011). Thus, in [Table 1], we see a shift from cell-based (HEp2-HeLa) to molecular-based (pCVD432) assays. [4]

Meta-analysis is an attractive and cheaper alternative to the primary study, which when large, is bound to be expensive and logistically problematic. Done rigorously, meta-analysis effects merit higher confidence and greater statistical precision. Its findings can unmask large-scale patterns not obvious in primary studies. Meta-analysis can then facilitate critical transfer of knowledge from researcher to clinician enabling analyses of important patient subgroups, delineation of high risk factors for infection enough for information to be useful for public health advice in risk for infection. Consequently, meta-analysis lends rigor to better assist health authorities in directing therapeutic decisions to target populations, urgency for health education and control measures. Indeed, in the public health domain, a number of difficult issues that had been repeatedly studied were either resolved or clarified by the application of meta-analysis techniques. [25] This has led some government guidelines to recommend meta-analysis as the preferred method of summarising evidence of effectiveness and safety of health technologies in the face of multiple study results. [26]

 Survey of Meta-Analyses Publications in Microbiology



To obtain a sense and estimate extent of the use of meta-analysis in microbiology, we surveyed PubMed literature (April 11, 2013) using the search terms, "meta-analysis" in the title box and "microbiology" in the all fields box. This resulted in 655 citations, 48 of which were excluded because they were not meta-analyses. We then categorised the remaining 607 according to country and three periods (1992-1999, 2000-2006 and 2007-2013). We counted the number of published meta-analyses from two, 13 and 11 countries grouped under North America, Europe and Asia, respectively and divided the totals by the number of years in each period, producing a meta-analysis output per year value.

[Figure 7] shows that use of meta-analysis research has grown dramatically in the last 6 years with Europe slightly ahead of the other two continents. The remarkable aspect of this growth is that of Asia, from a miniscule output in the 90s to a level comparable with that of North America 15 years later. The Asian and North American 2007-2013 values were attributed to China and the USA, respectively, both pegged at 77%, demonstrating the dominance of these countries in term of meta-analysis publications.{Figure 7}

 Conclusions



Microbiology research outputs, as in other biomedical disciplines, are increasing at an exponential rate. Because of the critical importance of microbiology findings across populations and geographical regions, objective evaluation of these primary studies may facilitate decision-making that impacts upon public health policy. Clinicians and researchers in this field would likely benefit from the use and interpretation of this statistical technique. The survey in this review delineates the increasing use of meta-analysis in microbiology in North America and Asia, where China asserted its growing capability to utilise output from primary studies microbiology research only recently.

In the EAEC meta-analysis example we cited in this review, we sought to examine the question of whether children with diarrhoea were more likely than those without diarrhoea to carry EAEC in stool (as detected by HEp2-HeLa and pCVD432 assays). The reason we thought this necessary was that the existing studies had such conflicting results that varied in geography, methodology, and sample size-leading to perceived controversy or uncertainty about the pathogenic nature of this organism. As our results show, the application of meta-analytic statistical approaches allowed us to confirm that EAEC is in fact a cause of acute diarrhoea in childhood, which was long considered controversial due to conflicting results from individual studies. Similar meta-analyses in microbiology may have the ability to advance our understanding of other emerging and established pathogens.

References

1Lewis D, Newton DC, Guy RJ, Ali H, Chen MY, Fairley CK, et al. The prevalence of Chlamydia trachomatis infection in Australia: A systematic review and meta-analysis. BMC Infect Dis 2012;12:113.
2Peralta-Rodríguez R, Romero-Morelos P, Villegas-Ruíz V, Mendoza-Rodríguez M, Taniguchi-Ponciano K, González-Yebra B, et al. Prevalence of human papillomavirus in the cervical epithelium of Mexican women: Meta-analysis. Infect Agents Cancer 2012;7:34.
3Beijer U, Wolf A, Fazel S. Prevalence of tuberculosis, hepatitis C virus, and HIV in homeless people: A systematic review and meta-analysis. Lancet Infect Dis 2012;12:859-70.
4Pabalan N, Singian E, Jarjanazi H, Steiner TS. Enteroaggregative Escherichia coli and acute diarrhea in children: A meta-analysis of South Asian populations. Eur J Clin Microbiol Infect Dis 2013;32:597-607.
5Yuan Y, Hunt RH. Systematic reviews: The good, the bad, and the ugly. Am J Gastroenterol 2009;104:1086-92.
6Rowland DY, Debanne SM. Meta-analysis, part I. Gastrointest Endosc 2002;55:612-3.
7Gotzsche PC. Why we need a broad perspective on meta-analysis. It may be crucially important for patients. BMJ 2000;321:585-6.
8Liberati A, Altman DG, Tetzlaff J, Mulrow C, Gøtzsche PC, Ioannidis JP, et al. The PRISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: Explanation and elaboration. J Clin Epidemiol 2009;62:e1-34.
9Lemeshow AR, Blum RE, Berlin JA, Stoto MA, Colditz GA. Searching one or two databases was insufficient for meta-analysis of observational studies. J Clin Epidemiol 2005;58:867-73.
10Egger M, Davey Smith G, Schneider M, Minder C. Bias in meta-analysis detected by a simple, graphical test. BMJ 1997;315:629-34.
11Ballal M, Shivananda PG. Rotavirus and enteric pathogens in infantile diarrhoea in Manipal, South India. Indian J Pediatr 2002;69:393-6.
12Rajendran P, Ajjampur SS, Chidambaram D, Chandrabose G, Thangaraj B, Sarkar R, et al. Pathotypes of diarrheagenic Escherichia coli in children attending a tertiary care hospital in South India. Diagn Microbiol Infect Dis 2010;68:117-22.
13Anvikar AR, Dolla C, Dutta S, Rao VG, Gadge VS, Shukla GP, et al. Role of Escherichia coli in acute diarrhoea in tribal preschool children of central India, Paediatr Perinat Epidemiol 2008;22:40-6.
14Ghosh AR, Nair GB, Naik TN, Paul M, Pal SC, Sen D. Entero-adherent Escherichia coli is an important diarrhoeagenic agent in infants aged below 6 months in Calcutta, India. J Med Microbiol 1992;36:264-8.
15Kang G, Mathan MM, Mathan VI. Evaluation of a simplified HEp-2 cell adherence assay for Escherichia coli isolated from south Indian children with acute diarrhea and controls. J Clin Microbiol 1995;33:2204-5.
16Meng CY, Smith BL, Bodhidatta L, Richard SA, Vansith K, Thy B, et al. Etiology of diarrhea in young children and patterns of antibiotic resistance in Cambodia, Pediatr Infect Dis J 2011;30:331-5.
17Galbraith RF. A note on graphical presentation of estimated odds ratios from several clinical trials. Stat Med 1988;7:889-94.
18Dutta S, Pal S, Chakrabarti S, Dutta P, Manna B. Use of PCR to identify enteroaggregative Escherichia coli as an important cause of acute diarrhoea among children living in Calcutta, India. J Med Microbiol 1999;48:1011-6.
19Mantel N, Haenszel W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer Inst 1959;22:719-48.
20DerSimonian R, Laird N. Meta-analysis in clinical trials. Control Clin Trials 1986;7:177-88.
21Lau J, Ioannidis JP, Schmid CH. Quantitative synthesis in systematic reviews. Ann Intern Med 1997;127:820-6.
22Higgins JP, Thompson SG. Quantifying heterogeneity in a meta-analysis. Stat Med 2002;21:1539-58.
23Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, et al. Meta-analysis of observational studies in epidemiology: A proposal for reporting. Meta-analysis Of Observational Studies in Epidemiology (MOOSE) group. JAMA 2000;283:2008-12.
24Begg CB, Mazumdar M. Operating characteristics of a rank correlation test for publication bias. Biometrics 1994;50:1088-101.
25Louis TA, Fineberg HV, Mosteller F. Findings for public health from meta-analysis, Annu Rev Public Health 1985;6:1-20.
26NICE. Guide to the methods of technology appraisal. Available from: http://www.nice.org.uk/pdf 2008 [Last accessed on 2013 May].